Taking Stanford's Machine Learning course this fall has given me a couple of new analogies to explain the innovator's dilemma. Think of business strategy as the

Taking Stanford's Machine Learning course this fall has given me a couple of new analogies to explain the innovator's dilemma. Think of business strategy as the learning hypothesis, and the learning algorithm seeks to increase profits by minimizing the error between what value they capture and the available dollars in the market. Successful companies walk the surface, making stepwise reductions in this error rate. This often leads to a perfectly rational descent to a local optima. These local optima can be very deep and very profitable, which makes it even more difficult to turn attention to steps that will introduce loss and variance in order to find a more profitable (or global) optima. "Randomizing" the starting points (the effect of hundreds of startups in a market) can increase the likelihood of a better optima being found, as can large companies attempting the same with lots of small bets (e.g., 20% time) which then are cultivated using the proven learning algorithm (i.e., management approach.) True visionaries are able to hasten the process by scouting out where these new optima are likely to reside.